Transfer Learning
نقل التعلم
Definition
Reusing the knowledge of a model trained on a large task or dataset to speed up and improve learning on another, related task.
Definition
Transfer learning is a machine learning approach in which knowledge gained while solving a source task is reused to improve learning on a different but related target task.
Practical examples
import torch.nn as nn
from torchvision import models
model = models.resnet18(weights="IMAGENET1K_V1") # يحمّل أوزاناً مدربة مسبقاً
for p in model.parameters():
p.requires_grad = False # تجميد الطبقات
model.fc = nn.Linear(model.fc.in_features, 3) # رأس جديد لثلاث فئات
trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)
print(trainable)Expected output
1539
Scientific sources
References and scientific sources
-
Research paperA Survey on Transfer Learning A survey of transfer learning and its types.
-
Research paperHow transferable are features in deep neural networks? A study of how transferable features are across layers.
